1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation
5 #ifndef RTE_EXEC_ENV_LINUX
6 #error "KNI is not supported"
12 #include <sys/ioctl.h>
14 #include <rte_spinlock.h>
15 #include <rte_string_fns.h>
16 #include <rte_ethdev.h>
17 #include <rte_malloc.h>
20 #include <rte_memzone.h>
21 #include <rte_tailq.h>
22 #include <rte_rwlock.h>
23 #include <rte_eal_memconfig.h>
24 #include <rte_kni_common.h>
25 #include "rte_kni_fifo.h"
27 #define MAX_MBUF_BURST_NUM 32
29 /* Maximum number of ring entries */
30 #define KNI_FIFO_COUNT_MAX 1024
31 #define KNI_FIFO_SIZE (KNI_FIFO_COUNT_MAX * sizeof(void *) + \
32 sizeof(struct rte_kni_fifo))
34 #define KNI_REQUEST_MBUF_NUM_MAX 32
36 #define KNI_MEM_CHECK(cond, fail) do { if (cond) goto fail; } while (0)
38 #define KNI_MZ_NAME_FMT "kni_info_%s"
39 #define KNI_TX_Q_MZ_NAME_FMT "kni_tx_%s"
40 #define KNI_RX_Q_MZ_NAME_FMT "kni_rx_%s"
41 #define KNI_ALLOC_Q_MZ_NAME_FMT "kni_alloc_%s"
42 #define KNI_FREE_Q_MZ_NAME_FMT "kni_free_%s"
43 #define KNI_REQ_Q_MZ_NAME_FMT "kni_req_%s"
44 #define KNI_RESP_Q_MZ_NAME_FMT "kni_resp_%s"
45 #define KNI_SYNC_ADDR_MZ_NAME_FMT "kni_sync_%s"
47 TAILQ_HEAD(rte_kni_list, rte_tailq_entry);
49 static struct rte_tailq_elem rte_kni_tailq = {
52 EAL_REGISTER_TAILQ(rte_kni_tailq)
58 char name[RTE_KNI_NAMESIZE]; /**< KNI interface name */
59 uint16_t group_id; /**< Group ID of KNI devices */
60 uint32_t slot_id; /**< KNI pool slot ID */
61 struct rte_mempool *pktmbuf_pool; /**< pkt mbuf mempool */
62 unsigned int mbuf_size; /**< mbuf size */
64 const struct rte_memzone *m_tx_q; /**< TX queue memzone */
65 const struct rte_memzone *m_rx_q; /**< RX queue memzone */
66 const struct rte_memzone *m_alloc_q;/**< Alloc queue memzone */
67 const struct rte_memzone *m_free_q; /**< Free queue memzone */
69 struct rte_kni_fifo *tx_q; /**< TX queue */
70 struct rte_kni_fifo *rx_q; /**< RX queue */
71 struct rte_kni_fifo *alloc_q; /**< Allocated mbufs queue */
72 struct rte_kni_fifo *free_q; /**< To be freed mbufs queue */
74 const struct rte_memzone *m_req_q; /**< Request queue memzone */
75 const struct rte_memzone *m_resp_q; /**< Response queue memzone */
76 const struct rte_memzone *m_sync_addr;/**< Sync addr memzone */
78 /* For request & response */
79 struct rte_kni_fifo *req_q; /**< Request queue */
80 struct rte_kni_fifo *resp_q; /**< Response queue */
81 void *sync_addr; /**< Req/Resp Mem address */
83 struct rte_kni_ops ops; /**< operations for request */
87 KNI_REQ_NO_REGISTER = 0,
91 static void kni_free_mbufs(struct rte_kni *kni);
92 static void kni_allocate_mbufs(struct rte_kni *kni);
94 static volatile int kni_fd = -1;
96 /* Shall be called before any allocation happens */
98 rte_kni_init(unsigned int max_kni_ifaces __rte_unused)
100 if (rte_eal_iova_mode() != RTE_IOVA_PA) {
101 RTE_LOG(ERR, KNI, "KNI requires IOVA as PA\n");
105 /* Check FD and open */
107 kni_fd = open("/dev/" KNI_DEVICE, O_RDWR);
110 "Can not open /dev/%s\n", KNI_DEVICE);
118 static struct rte_kni *
119 __rte_kni_get(const char *name)
122 struct rte_tailq_entry *te;
123 struct rte_kni_list *kni_list;
125 kni_list = RTE_TAILQ_CAST(rte_kni_tailq.head, rte_kni_list);
127 TAILQ_FOREACH(te, kni_list, next) {
129 if (strncmp(name, kni->name, RTE_KNI_NAMESIZE) == 0)
140 kni_reserve_mz(struct rte_kni *kni)
142 char mz_name[RTE_MEMZONE_NAMESIZE];
144 snprintf(mz_name, RTE_MEMZONE_NAMESIZE, KNI_TX_Q_MZ_NAME_FMT, kni->name);
145 kni->m_tx_q = rte_memzone_reserve(mz_name, KNI_FIFO_SIZE, SOCKET_ID_ANY, 0);
146 KNI_MEM_CHECK(kni->m_tx_q == NULL, tx_q_fail);
148 snprintf(mz_name, RTE_MEMZONE_NAMESIZE, KNI_RX_Q_MZ_NAME_FMT, kni->name);
149 kni->m_rx_q = rte_memzone_reserve(mz_name, KNI_FIFO_SIZE, SOCKET_ID_ANY, 0);
150 KNI_MEM_CHECK(kni->m_rx_q == NULL, rx_q_fail);
152 snprintf(mz_name, RTE_MEMZONE_NAMESIZE, KNI_ALLOC_Q_MZ_NAME_FMT, kni->name);
153 kni->m_alloc_q = rte_memzone_reserve(mz_name, KNI_FIFO_SIZE, SOCKET_ID_ANY, 0);
154 KNI_MEM_CHECK(kni->m_alloc_q == NULL, alloc_q_fail);
156 snprintf(mz_name, RTE_MEMZONE_NAMESIZE, KNI_FREE_Q_MZ_NAME_FMT, kni->name);
157 kni->m_free_q = rte_memzone_reserve(mz_name, KNI_FIFO_SIZE, SOCKET_ID_ANY, 0);
158 KNI_MEM_CHECK(kni->m_free_q == NULL, free_q_fail);
160 snprintf(mz_name, RTE_MEMZONE_NAMESIZE, KNI_REQ_Q_MZ_NAME_FMT, kni->name);
161 kni->m_req_q = rte_memzone_reserve(mz_name, KNI_FIFO_SIZE, SOCKET_ID_ANY, 0);
162 KNI_MEM_CHECK(kni->m_req_q == NULL, req_q_fail);
164 snprintf(mz_name, RTE_MEMZONE_NAMESIZE, KNI_RESP_Q_MZ_NAME_FMT, kni->name);
165 kni->m_resp_q = rte_memzone_reserve(mz_name, KNI_FIFO_SIZE, SOCKET_ID_ANY, 0);
166 KNI_MEM_CHECK(kni->m_resp_q == NULL, resp_q_fail);
168 snprintf(mz_name, RTE_MEMZONE_NAMESIZE, KNI_SYNC_ADDR_MZ_NAME_FMT, kni->name);
169 kni->m_sync_addr = rte_memzone_reserve(mz_name, KNI_FIFO_SIZE, SOCKET_ID_ANY, 0);
170 KNI_MEM_CHECK(kni->m_sync_addr == NULL, sync_addr_fail);
175 rte_memzone_free(kni->m_resp_q);
177 rte_memzone_free(kni->m_req_q);
179 rte_memzone_free(kni->m_free_q);
181 rte_memzone_free(kni->m_alloc_q);
183 rte_memzone_free(kni->m_rx_q);
185 rte_memzone_free(kni->m_tx_q);
191 kni_release_mz(struct rte_kni *kni)
193 rte_memzone_free(kni->m_tx_q);
194 rte_memzone_free(kni->m_rx_q);
195 rte_memzone_free(kni->m_alloc_q);
196 rte_memzone_free(kni->m_free_q);
197 rte_memzone_free(kni->m_req_q);
198 rte_memzone_free(kni->m_resp_q);
199 rte_memzone_free(kni->m_sync_addr);
203 rte_kni_alloc(struct rte_mempool *pktmbuf_pool,
204 const struct rte_kni_conf *conf,
205 struct rte_kni_ops *ops)
208 struct rte_kni_device_info dev_info;
210 struct rte_tailq_entry *te;
211 struct rte_kni_list *kni_list;
213 if (!pktmbuf_pool || !conf || !conf->name[0])
216 /* Check if KNI subsystem has been initialized */
218 RTE_LOG(ERR, KNI, "KNI subsystem has not been initialized. Invoke rte_kni_init() first\n");
222 rte_mcfg_tailq_write_lock();
224 kni = __rte_kni_get(conf->name);
226 RTE_LOG(ERR, KNI, "KNI already exists\n");
230 te = rte_zmalloc("KNI_TAILQ_ENTRY", sizeof(*te), 0);
232 RTE_LOG(ERR, KNI, "Failed to allocate tailq entry\n");
236 kni = rte_zmalloc("KNI", sizeof(struct rte_kni), RTE_CACHE_LINE_SIZE);
238 RTE_LOG(ERR, KNI, "KNI memory allocation failed\n");
242 strlcpy(kni->name, conf->name, RTE_KNI_NAMESIZE);
245 memcpy(&kni->ops, ops, sizeof(struct rte_kni_ops));
247 kni->ops.port_id = UINT16_MAX;
249 memset(&dev_info, 0, sizeof(dev_info));
250 dev_info.core_id = conf->core_id;
251 dev_info.force_bind = conf->force_bind;
252 dev_info.group_id = conf->group_id;
253 dev_info.mbuf_size = conf->mbuf_size;
254 dev_info.mtu = conf->mtu;
255 dev_info.min_mtu = conf->min_mtu;
256 dev_info.max_mtu = conf->max_mtu;
258 memcpy(dev_info.mac_addr, conf->mac_addr, RTE_ETHER_ADDR_LEN);
260 strlcpy(dev_info.name, conf->name, RTE_KNI_NAMESIZE);
262 ret = kni_reserve_mz(kni);
267 kni->tx_q = kni->m_tx_q->addr;
268 kni_fifo_init(kni->tx_q, KNI_FIFO_COUNT_MAX);
269 dev_info.tx_phys = kni->m_tx_q->phys_addr;
272 kni->rx_q = kni->m_rx_q->addr;
273 kni_fifo_init(kni->rx_q, KNI_FIFO_COUNT_MAX);
274 dev_info.rx_phys = kni->m_rx_q->phys_addr;
277 kni->alloc_q = kni->m_alloc_q->addr;
278 kni_fifo_init(kni->alloc_q, KNI_FIFO_COUNT_MAX);
279 dev_info.alloc_phys = kni->m_alloc_q->phys_addr;
282 kni->free_q = kni->m_free_q->addr;
283 kni_fifo_init(kni->free_q, KNI_FIFO_COUNT_MAX);
284 dev_info.free_phys = kni->m_free_q->phys_addr;
287 kni->req_q = kni->m_req_q->addr;
288 kni_fifo_init(kni->req_q, KNI_FIFO_COUNT_MAX);
289 dev_info.req_phys = kni->m_req_q->phys_addr;
292 kni->resp_q = kni->m_resp_q->addr;
293 kni_fifo_init(kni->resp_q, KNI_FIFO_COUNT_MAX);
294 dev_info.resp_phys = kni->m_resp_q->phys_addr;
296 /* Req/Resp sync mem area */
297 kni->sync_addr = kni->m_sync_addr->addr;
298 dev_info.sync_va = kni->m_sync_addr->addr;
299 dev_info.sync_phys = kni->m_sync_addr->phys_addr;
301 kni->pktmbuf_pool = pktmbuf_pool;
302 kni->group_id = conf->group_id;
303 kni->mbuf_size = conf->mbuf_size;
305 ret = ioctl(kni_fd, RTE_KNI_IOCTL_CREATE, &dev_info);
311 kni_list = RTE_TAILQ_CAST(rte_kni_tailq.head, rte_kni_list);
312 TAILQ_INSERT_TAIL(kni_list, te, next);
314 rte_mcfg_tailq_write_unlock();
316 /* Allocate mbufs and then put them into alloc_q */
317 kni_allocate_mbufs(kni);
328 rte_mcfg_tailq_write_unlock();
334 kni_free_fifo(struct rte_kni_fifo *fifo)
337 struct rte_mbuf *pkt;
340 ret = kni_fifo_get(fifo, (void **)&pkt, 1);
342 rte_pktmbuf_free(pkt);
347 va2pa(struct rte_mbuf *m)
349 return (void *)((unsigned long)m -
350 ((unsigned long)m->buf_addr -
351 (unsigned long)m->buf_iova));
355 va2pa_all(struct rte_mbuf *mbuf)
357 void *phy_mbuf = va2pa(mbuf);
358 struct rte_mbuf *next = mbuf->next;
360 mbuf->next = va2pa(next);
368 obj_free(struct rte_mempool *mp __rte_unused, void *opaque, void *obj,
369 unsigned obj_idx __rte_unused)
371 struct rte_mbuf *m = obj;
372 void *mbuf_phys = opaque;
374 if (va2pa(m) == mbuf_phys)
379 kni_free_fifo_phy(struct rte_mempool *mp, struct rte_kni_fifo *fifo)
385 ret = kni_fifo_get(fifo, &mbuf_phys, 1);
387 rte_mempool_obj_iter(mp, obj_free, mbuf_phys);
392 rte_kni_release(struct rte_kni *kni)
394 struct rte_tailq_entry *te;
395 struct rte_kni_list *kni_list;
396 struct rte_kni_device_info dev_info;
402 kni_list = RTE_TAILQ_CAST(rte_kni_tailq.head, rte_kni_list);
404 rte_mcfg_tailq_write_lock();
406 TAILQ_FOREACH(te, kni_list, next) {
414 strlcpy(dev_info.name, kni->name, sizeof(dev_info.name));
415 if (ioctl(kni_fd, RTE_KNI_IOCTL_RELEASE, &dev_info) < 0) {
416 RTE_LOG(ERR, KNI, "Fail to release kni device\n");
420 TAILQ_REMOVE(kni_list, te, next);
422 rte_mcfg_tailq_write_unlock();
424 /* mbufs in all fifo should be released, except request/response */
426 /* wait until all rxq packets processed by kernel */
427 while (kni_fifo_count(kni->rx_q) && retry--)
430 if (kni_fifo_count(kni->rx_q))
431 RTE_LOG(ERR, KNI, "Fail to free all Rx-q items\n");
433 kni_free_fifo_phy(kni->pktmbuf_pool, kni->alloc_q);
434 kni_free_fifo(kni->tx_q);
435 kni_free_fifo(kni->free_q);
446 rte_mcfg_tailq_write_unlock();
451 /* default callback for request of configuring device mac address */
453 kni_config_mac_address(uint16_t port_id, uint8_t mac_addr[])
457 if (!rte_eth_dev_is_valid_port(port_id)) {
458 RTE_LOG(ERR, KNI, "Invalid port id %d\n", port_id);
462 RTE_LOG(INFO, KNI, "Configure mac address of %d", port_id);
464 ret = rte_eth_dev_default_mac_addr_set(port_id,
465 (struct rte_ether_addr *)mac_addr);
467 RTE_LOG(ERR, KNI, "Failed to config mac_addr for port %d\n",
473 /* default callback for request of configuring promiscuous mode */
475 kni_config_promiscusity(uint16_t port_id, uint8_t to_on)
479 if (!rte_eth_dev_is_valid_port(port_id)) {
480 RTE_LOG(ERR, KNI, "Invalid port id %d\n", port_id);
484 RTE_LOG(INFO, KNI, "Configure promiscuous mode of %d to %d\n",
488 ret = rte_eth_promiscuous_enable(port_id);
490 ret = rte_eth_promiscuous_disable(port_id);
494 "Failed to %s promiscuous mode for port %u: %s\n",
495 to_on ? "enable" : "disable", port_id,
501 /* default callback for request of configuring allmulticast mode */
503 kni_config_allmulticast(uint16_t port_id, uint8_t to_on)
505 if (!rte_eth_dev_is_valid_port(port_id)) {
506 RTE_LOG(ERR, KNI, "Invalid port id %d\n", port_id);
510 RTE_LOG(INFO, KNI, "Configure allmulticast mode of %d to %d\n",
514 rte_eth_allmulticast_enable(port_id);
516 rte_eth_allmulticast_disable(port_id);
522 rte_kni_handle_request(struct rte_kni *kni)
525 struct rte_kni_request *req = NULL;
530 /* Get request mbuf */
531 ret = kni_fifo_get(kni->req_q, (void **)&req, 1);
533 return 0; /* It is OK of can not getting the request mbuf */
535 if (req != kni->sync_addr) {
536 RTE_LOG(ERR, KNI, "Wrong req pointer %p\n", req);
540 /* Analyze the request and call the relevant actions for it */
541 switch (req->req_id) {
542 case RTE_KNI_REQ_CHANGE_MTU: /* Change MTU */
543 if (kni->ops.change_mtu)
544 req->result = kni->ops.change_mtu(kni->ops.port_id,
547 case RTE_KNI_REQ_CFG_NETWORK_IF: /* Set network interface up/down */
548 if (kni->ops.config_network_if)
549 req->result = kni->ops.config_network_if(kni->ops.port_id,
552 case RTE_KNI_REQ_CHANGE_MAC_ADDR: /* Change MAC Address */
553 if (kni->ops.config_mac_address)
554 req->result = kni->ops.config_mac_address(
555 kni->ops.port_id, req->mac_addr);
556 else if (kni->ops.port_id != UINT16_MAX)
557 req->result = kni_config_mac_address(
558 kni->ops.port_id, req->mac_addr);
560 case RTE_KNI_REQ_CHANGE_PROMISC: /* Change PROMISCUOUS MODE */
561 if (kni->ops.config_promiscusity)
562 req->result = kni->ops.config_promiscusity(
563 kni->ops.port_id, req->promiscusity);
564 else if (kni->ops.port_id != UINT16_MAX)
565 req->result = kni_config_promiscusity(
566 kni->ops.port_id, req->promiscusity);
568 case RTE_KNI_REQ_CHANGE_ALLMULTI: /* Change ALLMULTICAST MODE */
569 if (kni->ops.config_allmulticast)
570 req->result = kni->ops.config_allmulticast(
571 kni->ops.port_id, req->allmulti);
572 else if (kni->ops.port_id != UINT16_MAX)
573 req->result = kni_config_allmulticast(
574 kni->ops.port_id, req->allmulti);
577 RTE_LOG(ERR, KNI, "Unknown request id %u\n", req->req_id);
578 req->result = -EINVAL;
582 /* Construct response mbuf and put it back to resp_q */
583 ret = kni_fifo_put(kni->resp_q, (void **)&req, 1);
585 RTE_LOG(ERR, KNI, "Fail to put the muf back to resp_q\n");
586 return -1; /* It is an error of can't putting the mbuf back */
593 rte_kni_tx_burst(struct rte_kni *kni, struct rte_mbuf **mbufs, unsigned int num)
595 num = RTE_MIN(kni_fifo_free_count(kni->rx_q), num);
596 void *phy_mbufs[num];
600 for (i = 0; i < num; i++)
601 phy_mbufs[i] = va2pa_all(mbufs[i]);
603 ret = kni_fifo_put(kni->rx_q, phy_mbufs, num);
605 /* Get mbufs from free_q and then free them */
612 rte_kni_rx_burst(struct rte_kni *kni, struct rte_mbuf **mbufs, unsigned int num)
614 unsigned int ret = kni_fifo_get(kni->tx_q, (void **)mbufs, num);
616 /* If buffers removed, allocate mbufs and then put them into alloc_q */
618 kni_allocate_mbufs(kni);
624 kni_free_mbufs(struct rte_kni *kni)
627 struct rte_mbuf *pkts[MAX_MBUF_BURST_NUM];
629 ret = kni_fifo_get(kni->free_q, (void **)pkts, MAX_MBUF_BURST_NUM);
630 if (likely(ret > 0)) {
631 for (i = 0; i < ret; i++)
632 rte_pktmbuf_free(pkts[i]);
637 kni_allocate_mbufs(struct rte_kni *kni)
640 struct rte_mbuf *pkts[MAX_MBUF_BURST_NUM];
641 void *phys[MAX_MBUF_BURST_NUM];
644 RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, pool) !=
645 offsetof(struct rte_kni_mbuf, pool));
646 RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, buf_addr) !=
647 offsetof(struct rte_kni_mbuf, buf_addr));
648 RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, next) !=
649 offsetof(struct rte_kni_mbuf, next));
650 RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, data_off) !=
651 offsetof(struct rte_kni_mbuf, data_off));
652 RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, data_len) !=
653 offsetof(struct rte_kni_mbuf, data_len));
654 RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, pkt_len) !=
655 offsetof(struct rte_kni_mbuf, pkt_len));
656 RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, ol_flags) !=
657 offsetof(struct rte_kni_mbuf, ol_flags));
659 /* Check if pktmbuf pool has been configured */
660 if (kni->pktmbuf_pool == NULL) {
661 RTE_LOG(ERR, KNI, "No valid mempool for allocating mbufs\n");
665 allocq_free = (kni->alloc_q->read - kni->alloc_q->write - 1)
666 & (MAX_MBUF_BURST_NUM - 1);
667 for (i = 0; i < allocq_free; i++) {
668 pkts[i] = rte_pktmbuf_alloc(kni->pktmbuf_pool);
669 if (unlikely(pkts[i] == NULL)) {
671 RTE_LOG(ERR, KNI, "Out of memory\n");
674 phys[i] = va2pa(pkts[i]);
677 /* No pkt mbuf allocated */
681 ret = kni_fifo_put(kni->alloc_q, phys, i);
683 /* Check if any mbufs not put into alloc_q, and then free them */
684 if (ret >= 0 && ret < i && ret < MAX_MBUF_BURST_NUM) {
687 for (j = ret; j < i; j++)
688 rte_pktmbuf_free(pkts[j]);
693 rte_kni_get(const char *name)
697 if (name == NULL || name[0] == '\0')
700 rte_mcfg_tailq_read_lock();
702 kni = __rte_kni_get(name);
704 rte_mcfg_tailq_read_unlock();
710 rte_kni_get_name(const struct rte_kni *kni)
715 static enum kni_ops_status
716 kni_check_request_register(struct rte_kni_ops *ops)
718 /* check if KNI request ops has been registered*/
720 return KNI_REQ_NO_REGISTER;
722 if (ops->change_mtu == NULL
723 && ops->config_network_if == NULL
724 && ops->config_mac_address == NULL
725 && ops->config_promiscusity == NULL
726 && ops->config_allmulticast == NULL)
727 return KNI_REQ_NO_REGISTER;
729 return KNI_REQ_REGISTERED;
733 rte_kni_register_handlers(struct rte_kni *kni, struct rte_kni_ops *ops)
735 enum kni_ops_status req_status;
738 RTE_LOG(ERR, KNI, "Invalid KNI request operation.\n");
743 RTE_LOG(ERR, KNI, "Invalid kni info.\n");
747 req_status = kni_check_request_register(&kni->ops);
748 if (req_status == KNI_REQ_REGISTERED) {
749 RTE_LOG(ERR, KNI, "The KNI request operation has already registered.\n");
753 memcpy(&kni->ops, ops, sizeof(struct rte_kni_ops));
758 rte_kni_unregister_handlers(struct rte_kni *kni)
761 RTE_LOG(ERR, KNI, "Invalid kni info.\n");
765 memset(&kni->ops, 0, sizeof(struct rte_kni_ops));
771 rte_kni_update_link(struct rte_kni *kni, unsigned int linkup)
775 const char *new_carrier;
782 snprintf(path, sizeof(path), "/sys/devices/virtual/net/%s/carrier",
785 fd = open(path, O_RDWR);
787 RTE_LOG(ERR, KNI, "Failed to open file: %s.\n", path);
791 ret = read(fd, old_carrier, 2);
796 old_linkup = (old_carrier[0] == '1');
798 new_carrier = linkup ? "1" : "0";
799 ret = write(fd, new_carrier, 1);
801 RTE_LOG(ERR, KNI, "Failed to write file: %s.\n", path);